Agentic Commerce: $1T Market, 3x Failure Rate, and What Comes Next

Walmart just pulled the plug on buying things inside ChatGPT. And the data behind the decision is brutal.

Conversion rates for products sold through OpenAI’s Instant Checkout — the buy-now buttons that lived right inside ChatGPT — were three times worse than when shoppers clicked out to Walmart’s own website. Not slightly worse. Three times.

That’s not a rounding error. That’s a product that doesn’t work.

This matters because Walmart wasn’t some small test case. They put 200,000 products into ChatGPT’s shopping experience. They ran it for months. And the conclusion was clear: people don’t want to buy things inside a chatbot. At least not the way OpenAI built it.

What Actually Happened

In October 2025, OpenAI launched “Instant Checkout” — a feature that let ChatGPT users browse products and buy them without ever leaving the chat. Walmart was the flagship partner, along with Shopify merchants and other retailers.

The pitch was simple: ask ChatGPT for product recommendations, see results, click buy, done. No switching tabs. No navigating a retailer’s website. One seamless flow.

It flopped.

Daniel Danker, Walmart’s executive VP, told CNBC that conversion rates inside ChatGPT were three times lower than their own website. And the problems went deeper than just conversions:

• No tax collection system. As of February 2026, OpenAI hadn’t built infrastructure for collecting and remitting state sales taxes in the US.
• No loyalty program integration. Walmart shoppers couldn’t use their Walmart+ benefits or earn rewards.
• No returns handling. Refunds, cancellations, and customer service all remained the merchant’s responsibility, but the checkout sat inside OpenAI’s interface.
• An accountability gap. When something went wrong with an order, whose problem was it? OpenAI’s? Walmart’s? Nobody had a good answer.
• Accuracy problems everywhere. Early partners reported wrong prices, incorrect product descriptions, and lagging inventory data. If the AI couldn’t see real-time stock and pricing, it hallucinated the shopping funnel — showing items that were out of stock or priced differently than expected.

And the scale of adoption was telling. Despite the splashy launch, only about 30 Shopify merchants ever fully integrated with Instant Checkout. OpenAI’s own internal teams reportedly described users as “high-intent browsers with low conversion rates” — people who were curious enough to ask ChatGPT for product recommendations but not trusting enough to actually buy.

Shopify confirmed they’re pulling out of native checkout inside ChatGPT entirely. Etsy is building its own dedicated ChatGPT app instead. The entire model of “buy things inside the chatbot” is being abandoned by the companies that tried it first.

What’s Replacing It

Starting the week of March 25, 2026, Walmart’s own AI chatbot — called Sparky — will begin operating inside ChatGPT. And later, inside Google’s Gemini.

This is a fundamentally different approach. Instead of OpenAI handling the checkout, Walmart’s own technology handles everything. Items you add through Sparky in ChatGPT sync with your Walmart app cart and your Walmart.com cart. Checkout happens through Walmart, with Walmart’s loyalty programs, Walmart’s tax calculation, and Walmart’s customer service.

Gap is doing something similar — launching its own shopping experience inside Gemini using Google’s new Universal Commerce Protocol.

Sephora is building a ChatGPT app that integrates loyalty programs and personalized recommendations, with in-app checkout planned for later.

The pattern is clear: retailers aren’t leaving AI shopping. They’re taking back control of it.

What Is Agentic Commerce, Anyway?

If you keep seeing this term everywhere and aren’t sure what it means — you’re not alone. It’s the biggest buzzword in retail right now, and the definition keeps shifting.

Agentic commerce is when AI agents don’t just recommend products — they actually help complete purchases on your behalf, within boundaries you set. You tell the AI what you need, and it handles the rest: finding options, comparing prices, checking availability, and eventually buying.

McKinsey defines it as “AI that anticipates consumer needs, navigates shopping options, negotiates deals, and executes transactions.”

In practice right now, it looks like this:

• You ask ChatGPT “I need running shoes under $120 with good arch support” and it shows you options from multiple retailers
• Google’s shopping agent can take a handwritten recipe, create a shopping list, and purchase the ingredients
• Nordstrom launched an AI agent before Black Friday that handled customer service inquiries
• A footwear retailer’s agent lets customers photograph damaged shoes, processes the claim instantly, and ships a replacement within 30 seconds

The market is real. According to IBM’s January 2026 research, 45% of consumers already use AI for at least part of their buying journey. During Cyber Week 2025, one in five orders involved an AI agent — roughly $70 billion in gross merchandise value.

McKinsey projects the US B2C retail opportunity for agentic commerce at $900 billion to $1 trillion by 2030, with global opportunities reaching $3 to $5 trillion.

So no — agentic commerce isn’t hype. But as Walmart just proved, the execution matters enormously.

Why Instant Checkout Failed (And What It Tells Us)

The Walmart failure isn’t just a product story. It reveals something important about how people actually shop.

Checkout isn’t just a button. It’s where loyalty programs, bundle discounts, upsell mechanics, returns policies, tax compliance, fraud detection, and customer identity all converge. When OpenAI tried to own that step, they stripped out everything that makes checkout work for both retailers and customers.

Trust lives with the brand, not the chatbot. A Forrester March 2026 survey confirmed that among US, UK, and Canadian adults who regularly use AI tools, completing a purchase inside the AI is their least adopted use case. People will happily ask AI for recommendations. They don’t want to hand it their credit card inside a chat window. Shoppers want to see the actual product page, read reviews, combine shipping with other items in their cart, and collect loyalty points. A chat interface strips all of that away.

Discovery and checkout are different jobs. AI is genuinely good at helping you find the right product — understanding vague requests, comparing options, filtering by your actual needs. But the moment you’re ready to buy, you want the retailer’s infrastructure: saved addresses, payment methods, order tracking, easy returns. As one ecommerce analyst put it: AI that removes friction wins, but AI that removes trust loses.

This is why the industry is splitting AI shopping into two layers:

1. Discovery — handled by AI (ChatGPT, Gemini, Perplexity)
2. Checkout — handled by retailers (Walmart, Shopify, Sephora, etc.)

The New Infrastructure: Google’s Universal Commerce Protocol

Behind the scenes, a more interesting story is developing. Google announced the Universal Commerce Protocol (UCP) at NRF 2026 on January 11 — an open standard designed to be the plumbing for agentic commerce.

UCP is co-developed with Shopify, Etsy, Wayfair, Target, and Walmart, and endorsed by over 20 global partners including Adyen, American Express, Best Buy, Flipkart, Macy’s, Mastercard, Stripe, The Home Depot, Visa, and Zalando.

What it does:

• Creates a standard way for AI agents to browse products, add items to carts, and trigger checkout — across any retailer
• Supports identity linking so your loyalty programs work across AI surfaces
• Integrates with Agent2Agent (A2A) protocol and Model Context Protocol (MCP) for AI-to-AI communication
• Lets shoppers save multiple items to a cart at once, like normal shopping

Think of it like this: UCP is trying to be the “HTTPS of shopping” — a standard protocol that any AI agent can use to interact with any retailer. If it succeeds, you could ask Claude, ChatGPT, or Gemini to buy something, and they’d all use the same underlying protocol to complete the transaction through the retailer’s own system.

What This Means for Regular People

If you’re using AI tools for work, creative projects, or learning (which, if you’re reading this, you probably are), here’s what the agentic commerce shift means for you:

AI shopping recommendations are about to get much better. The move from “buy inside the chatbot” to “discover with AI, buy from the retailer” means retailers are investing heavily in making their products discoverable by AI agents. More structured data, better product descriptions, real-time inventory — all of which makes AI product recommendations more accurate.

Your existing accounts and loyalty programs will work with AI. The Walmart Sparky approach means your Walmart+ benefits, saved addresses, and order history all work when you shop through AI. Same for Sephora loyalty points, Gap rewards, etc.

Don’t buy things through AI checkout yet. Seriously. If any AI tool offers to process your payment directly inside the chat — without redirecting to the retailer — think twice. The tax, returns, and customer service infrastructure isn’t there yet. Walmart proved that with 200,000 products and millions of dollars.

AI is better at shopping research than shopping. Right now, the best use of AI in commerce is asking it to compare products, check reviews, find deals, and narrow down options. Let the AI do the thinking, then buy through the retailer directly.

The Numbers

The Bigger Picture

Agentic commerce isn’t dead. Walmart’s checkout inside ChatGPT is.

The difference matters. The vision — AI agents that help you shop smarter, faster, and more personally — is backed by real data and real money. McKinsey’s trillion-dollar projections, IBM’s 45% adoption numbers, and the 20+ companies building UCP infrastructure aren’t going away.

But the first attempt at execution — letting AI platforms own the checkout — failed for the exact reasons you’d expect if you’ve ever tried to return something bought through an unfamiliar interface.

The next wave looks different. Retailers own the transaction. AI owns the discovery. And protocols like UCP make them work together seamlessly.

But there’s a deeper question nobody has answered yet: when an AI agent shops on your behalf, how does the merchant verify it’s actually you? How does consent work? Who’s liable if an agent buys the wrong thing? These identity and accountability problems are the real bottleneck — not the technology itself. Gartner predicts 40% of agentic AI projects will fail by 2027 due to poor risk management and unclear ROI, and the identity question is a big reason why.

It’s messier than the “buy everything in one chat” dream. But it’s the version that might actually work.

Walmart’s Instant Checkout failure was reported March 20-24, 2026. Sparky integration with ChatGPT begins the week of March 25, 2026.

Sources:

• CNBC — OpenAI’s first try at agentic shopping stumbled
• Search Engine Land — Walmart: ChatGPT checkout converted 3x worse
• PYMNTS — Walmart Shuts Down Agentic Commerce With OpenAI
• Axios — Walmart, Gap test AI shopping in ChatGPT and Gemini
• McKinsey — The Agentic Commerce Opportunity
• Google Developers — Universal Commerce Protocol
• Forrester — Predictions 2026: The Agentic Commerce Race
• Stripe — Three biggest agentic commerce trends from NRF 2026

Related courses: AI Fundamentals | Prompt Engineering